V2 verses V3 quality

[Dave Gustafson]

I have extreme respect, admiration for the Firefly team and engineers, but something has gone off the rails concerning the V3 preview. This isn't a question, so I placed it in this section of bugs. If I didn't care about Firefly I wouldn't be here showing examples.

These are direct renders of "multiple generations" using V3 preview off the original V2 file. I will show the V2 first, then V3 will follow each image. I kept trying to get a good image from V3, yes it's a preview, but it needs to be addressed. These are the best of the best V3 images. And with that said I'll shut up lol and let you judge.

V2

V3

 

V2

V3

V2

V3

 

V2

 

V3

 

 

 

 

 

[rickg76305781]

In this example, the subject outputs are similar but the image 3 output appears to be "cleaner" and more of a refined vector illustartion or cartoon vs image 2 outputs which is more "raw". 

I expect some variation within each image group. I am applying 7 parameters varation is expected. BUT  as you can see. I tried multiple iterations with both image 2 and image 3. The results within each image group were generally consistant but were markedly different between 2 vs 3. raw vs refined. rough vs vector illuustration.

rickg

[@mj]

Perhaps an explanation on how and why the gen engines are decided upon.

Please consider leaving previous gen engines active.

mj

iMSD

[Amy27298162kmmg]

I agree. Please let us have access to Image 1 model again. In looking closer at my Image 2 model results, they are not as good as the Image 1 results. And the Image 3 results are not even usable for me. I loved the results when I first started using the Image 1. Now I'm losing interest.

[BlueJeanBaby]

Same prompt/styles, etc. here. V3's results - just awful. Embarrassing at best. I'm glad you're here in this thread, Kartika - I trust you'll pass along all this feedback. Thanks. 

[BlueJeanBaby]

Version 2 and 3 here, respectively. Same exact styles/prompts, etc.

[BlueJeanBaby]

What 3 generated is laughable and amateurish at best! 

 

[BlueJeanBaby]

Another example (please see this, and the following reply). Version 2 - cute little dragon on a skateboard...

 

[BlueJeanBaby]

. . . Then this (see prior reply). Version 3. I wonder if anyone has a worse example than this? Adobe - are you going to continue to allow this 'quality' with Model 3? Please keep V2 up! 

 

[droopydog500]

These strengthen my belief that from a realism perspective, 3>2 and from an artistic perspective 2>3...

    droopy

[BlueJeanBaby]

Agreed...somewhat. What realism though do those represent? lol  - But yes, in general, you're correct. 

[droopydog500]

I think those are a side effect of the realism push. Try to make everything look realistic.  That obviously does not always work out.

[BlueJeanBaby]

Most definitely. Why even have the 'art' capability then? Every prompt/style/formula worked before - no matter how many effects, themes, movements, concepts, etc. that I used. They've seriously limited the capabilities now in their quest for 'realism'. I've also noticed an Asian slant to many generation results, when nothing even was prompted as such. 

[Amy27298162kmmg]

Is it just me, or does Image 2 even seem worse than it used to be? 

[Mark DeMent]

It's not just you. Version 2 results recently have been much worse. But still better than 3. I can't use anything I've generated from 3. Everything looks like a bad Disney animation cell. I hope they return 2 to it's original state, and keep it.